PROJECT SUMMARY Phenomic advances from large-scale electronic health records (EHR) linked to DNA biobanks have pioneered an efficient approach to genetic discovery that has transformed human genetic studies, with the enormous potential to provide constraints on relevant biological mechanisms on a wide spectrum of human phenotypes. Nevertheless, our understanding of the downstream molecular consequences of genetic associations remains limited and impedes our ability to develop novel therapeutic strategies for complex diseases. Given their enormous discovery potential for human genomics and precision medicine, genetic analyses in diverse populations offer unprecedented opportunities to identify causal genetic mechanisms underlying human trait variation. This research proposal aims to address these convergent developments and critical gaps and to exert a powerful influence on efforts to expand our understanding of disease mechanisms and therapeutic possibilities. Here we hypothesize that a comprehensive multi- omic, phenomic, and trans-ethnic computational methodology will provide a robust and rigorous framework. This proposal thus has the following aims: AIM 1: Develop a regularized regression based methodology and a deep learning framework to improve characterization of the genetic architecture of gene expression and to build robust prediction models, extending a Transcriptome-Wide Association Study (TWAS) methodology (called PrediXcan) that we developed. AIM 2: Develop statistical causal modeling of trait-associated genetic variation through a convergent TWAS and Mendelian Randomization approach and apply it to thousands of human traits with available GWAS and EHR data. AIM 3: Develop analytic approaches and software tools to further genetic analyses in admixed and multi-ethnic populations and to lay the groundwork for trans-ethnic multi-omic methodologies, using EHR data (e.g., BioVU, UK Biobank, All of Us).